Method of and device for determining words in a speech signal

In a method of determining a sequence of words of a predetermined vocabulary in a speech signal, wherein the method includes the steps of:

• sampling the speech signal at recurrent instants so as to produce a sequence of test signals;
• comparing the test signals with different sequences of reference signals to generate scores, each sequence of reference signals representing a word of the vocabulary,
• commencing the comparison anew with successive test signals, starting from the beginning of sequences of reference signals;
• deriving, for each word end reached, a word result which comprises a reference to the beginning of the ended word, a word score, and a reference to the ended word;
• deriving at least one sequence of words from the word results,
• carrying out as far as the word end, and independently of any other comparison, each comparison that commences anew with the beginning of sequences of reference signals;
• forming a new preliminary word string by joining each word which ends at an instantaneous test signal to each preliminary word string which ends directly before the beginning of that word;
• combining the word score of each word with the score reached at the end of the preliminary word string so as to form a new score, most favorable new score being used as an initial score for a comparison newly started immediately thereafter, beginning with the different sequences of reference signals;
• and
• successively outputting the words of at least one word string from the beginning.